The evidence did not change. The model did.
The original report treated eight explanations as mutually exclusive. They were not. A late Trump decision, expected Israeli action and a regime-change opportunity can all describe the same sequence. Making them compete for one fixed pool of probability created a horse race that the evidence did not require.
The repaired model asks separate questions.
When was the decision made?
The 45-day military buildup and the 11 February UK basing request are somewhat more expected if the political decision had already been made. But contingency planning can produce both, and the rest of the government was strikingly unprepared.
The attack began before the force flow was complete. Bombers were not positioned forward. Britain refused basing on the night. Gulf hosts were not warned. There was no immediate sanctions package, legal letter, congressional funding request or public case for war ready to go.
Conditional on that preparation, the reported final sequence points the other way. Netanyahu supplies a perishable leadership-targeting opportunity. A small group meets. Vance opposes the attack. The advisers do not reach consensus. The decision is described as instinctive. The execute order goes out while diplomacy still appears live.
The preparation record gives 1.5 to 1 support for early commitment. The final sequence gives 4 to 1 support for a late decision. Combined, the evidence favours late commitment by about 2.7 to 1.
What influenced the decision?
The repaired analysis tests possible causes separately. The late targeting window, Israeli action and regime vulnerability each receive 4 to 1 evidential support. These are compatible, so their probabilities do not divide 100% and cannot be added together.
The official nuclear and diplomacy account and the narrow arms-money account each receive a 1 to 4 evidence factor. They are weakened, not eliminated.
Institutional narrowing and domestic political diversion each receive a factor of 1. The frozen public record does not move the starting view in either direction.
What has been withdrawn
The old 26.5%, 26.0%, 22.1% and 20.9% ranking is retained only as a historical record. The 53% versus 43% comparison is withdrawn, along with the claims that the official story, institutional momentum, arms-money and domestic diversion had been driven effectively to zero.
Members can also inspect the original source working, which remains available with a correction notice.